SME vs Large Firm AI Adoption Gap (OECD 2024)
Headline figures
| Firm size | Share using AI (OECD, 2024 or latest available) |
|---|---|
| Small (10–49 employees) | 11.9% |
| Medium (50–249) | 20.4% |
| Large (250+) | 40.0% |
Large-firm adoption is 3.4× small-firm adoption. Source figure: oecd-sme-ai-adoption-2025, Figure 2, p. 11 (OECD ICT Access and Usage Database, accessed July 2025).
Cross-country dispersion
The gap holds in every G7 country and in every OECD economy with comparable data. Country range observed in Figure 2:
- Korea, Denmark, Sweden, Belgium, Finland: ALL-firm adoption 25–30%, with large-firm rates of 55–70%.
- Germany, Austria, Norway: 20–25% all-firm, large-firm 50–60%.
- OECD average: 14% all-firm, ~40% large, 20.4% medium, 11.9% small.
- France: ~10% all-firm, ~30% large.
- Italy, UK, Canada: 6–9% all-firm.
- US, Japan, Australia, Israel: lowest in the OECD ICT survey (3–6% all-firm) — but the US figure understates because the underlying survey is 2021 and the methodology differs (see Annex).
Comparison to other digital tech
| Technology | Small/large adoption ratio |
|---|---|
| Cloud computing | ~1/2 |
| Internet of things | ~1/2 |
| AI | ~1/3.4 (less than one-third) |
The AI gap is materially wider than the gap for any other digital technology — a key data point because it suggests AI has distinct adoption barriers beyond the usual size-related digital divide.
Application-specific gaps (EU 2024, Eurostat)
| Application | Small (10–49) | Large (250+) | Ratio |
|---|---|---|---|
| Text mining | 5.76% | 21.44% | 3.7× |
| Natural language generation | 4.58% | 16.64% | 3.6× |
| Speech recognition | 3.92% | 16.70% | 4.3× |
| Machine learning (data analysis) | 3.13% | 20.58% | 6.6× |
| Workflow automation / decision support | 3.03% | 20.40% | 6.7× |
| Image recognition / processing | 2.60% | 13.40% | 5.2× |
| Autonomous robots / vehicles / drones | 0.66% | 7.19% | 10.9× |
Source: Eurostat 2025, Figure 3 in oecd-sme-ai-adoption-2025, p. 12. The gap scales with capital intensity — the more compute / specialized hardware / integration cost an application requires, the larger the SME-vs-large divergence.
Core-business adoption (G7, 2024)
AI adoption in core business functions (i.e., AI directly involved in producing goods or services, not peripheral admin / marketing):
| Country | Core-AI adoption share |
|---|---|
| Japan | 1.9% |
| … | … |
| United States | 6.1% |
Source: Filippucci et al. (2025), cited in oecd-sme-ai-adoption-2025 p. 9. Full G7 country range: 1.9% to 6.1%. Roughly an order of magnitude smaller than headline adoption shares because most current AI use is peripheral.
SME-internal: core vs peripheral
Among SMEs that do use generative AI:
- Only 29% use it in core activities (OECD 2025 SME survey).
- The majority use it for marketing, content drafting, customer service, internal documentation, research summarization.
Caveats
- Comparisons across countries should be made with caution — definitions of AI differ, sampling frames differ, and reference years range from 2020 (Israel, UK, Colombia) to 2024 (most EU members).
- The OECD ICT Access and Usage survey has a 10-employee floor — micro-firms (<10 employees), which are the bulk of SMEs by count, are not represented in these aggregate adoption shares.
- “AI use” is a binary survey response. It does not distinguish a Champion-level deployment from a single employee using ChatGPT.
Related
- oecd-sme-ai-adoption-2025 — primary source.
- sme-ai-adoption-gap — concept page.
- oecd-ai-adoption-trajectory-2020-2024 — the trend that contextualizes the gap.
- oecd-sectoral-ai-adoption-2024 — sector compounding factor.
- oecd-sme-adopter-taxonomy — what “AI use” disaggregates into.
- diffusion-vs-innovation — the literal substrate of the diffusion S-curve.
- H2_u-curve-of-value — bottom-right of the U is thin.
- scaling-gap — different gap, different denominator.